He is articulate about the craft as well, and puts its central trade-off better than most — too little direction and the work drifts, too much and it stops being what you meant.
Proktor
Jermaine Broadway · Lochlan Vaughan · Alan Guo · Andrew Needham
The problem they picked
Estimates of how many college students cheat run anywhere from 30% to 70%, and the team kept the whole range rather than reaching for the largest number. What redirected the project was a figure from the other side of the desk: roughly 40% of faculty who knew about cheating never reported it, because the process was involved or because they had nothing that would hold up. The problem is not that cheating is invisible. It is that an instructor who sees it has nothing to point at.
Proktor exists to produce that evidence — gaze tracking through the session, then a heatmap and a risk score an instructor can actually put in front of someone.
“A study came out that 50 to 70% of students in higher education cheat on exams. We built Proktor to solve that.”
Proktor, at the showcase
How it was built
Computer Vision Basics
Face-landmark tracking on the webcam feed estimates where on the screen a test-taker is actually looking.
Data Visualization
Those gaze samples accumulate into a heatmap laid over the exam page once the session ends.
Database Management
Supabase, with its own migrations, for assignments, session codes, and flagged students.
MediaPipe / face-apiComputer Vision
SupabaseBackend & Database
ViteBuild Tool
OpenRouterAI Model Gateway
Also in the buildWebcam gaze tracking · Post-session heatmap and risk score · Assignment builder and session codes
